AI로 악성앱 분석 시간 96% 단축… ‘AX 스파크 대상’ 강희도 님

SK Telecom held its internal AI agent competition ‘AX Spark’ in November, where 173 entries were narrowed to five finalists. The winner was Kang Hee-do of the Social Engineering Threat Response team for developing M-TripleA, an AI agent that automates malicious-app analysis. The system identifies command-and-control servers and flags infected devices for police intervention before financial harm occurs.

M-TripleA combines multiple AI agents that trace code and data flows, autonomously form and test hypotheses, and adjust strategies to detect malicious behavior and C2 addresses. The agent has reduced manual analysis workload from roughly 19 hours per app to about 50 minutes, a 96% time saving. Police have used its outputs to prevent ₩51.5 billion in phishing losses across 940 customers.

Kang Hee-do, who built the system alone, said the biggest technical challenge was maintaining coherent context among collaborating agents during long-running, iterative analysis. The project began in March after phishing losses nationwide surpassed ₩1.3 trillion, prompting SK Telecom and police to build a joint prevention framework.

Since deployment, analysts have shifted from manual tracing to exception handling and result validation. SK Telecom plans further optimization to cut inference costs while preserving detection accuracy. The company’s AX initiative is encouraging staff to embed AI directly into workflows rather than treating it as a generic productivity tool.